自動化花朵授粉微型飛行器之研發

近年來，蜂群衰竭失調症問題備受重視，同時授粉鳥類與哺乳類逐步面臨滅絕危機，有許多研究正在試圖釐清上述問題，亦有研究者開始尋找對應的授粉替代方案。而於溫網室設施中，授粉昆蟲無法進入，故需搭配人工授粉或購買授粉專用蜜蜂，然而人工授粉需要耗費人力、蜜蜂授粉效率高但仍存在一些限制。近幾年來的研究中，甚少有針對授粉飛行機器人的研究，尤其是仿昆蟲授粉、人工授粉等需要以實體元件接觸到花朵者，更是難以找尋。因此，為填補飛行器授粉自動化研究實作上的空缺，以及探討飛行器是否可能用於花朵授粉的可能性，本研究以四軸飛行器搭載影像模組與輕量化授粉機構，研發可應用於花朵授粉自動化之授粉機器人系統。授粉機器人系統可分為四部分，授粉機器人、影像定位系統、決策控制系統、花朵辨識系統。研究中使用深度攝影機定位飛行器，進行飛行器座標的自動控制，並透過影像模組尋找並抵達花朵正上方，再根據面積對高度的關係式回歸出目標花朵座標，最後完成以授粉物件碰觸花面的授粉作業。本研究中飛行器於定點飛行時，於七重複的條件下，其距離誤差RMSE（均方根誤差）之平均值於X、Y、Z軸分別為4.96公分、2.97公分、4.29公分，飛行器座標與目標位置之絕對距離的RMSE值之平均值則為7.27公分，其代表於飛行器的控制穩定性上已有相當的水準。而授粉試驗中，授粉機器人從切換至自動模式後，至最終完成棉球碰觸花面的授粉作業，一共花費了24.71秒。綜合以上，本研究成功的建立一套授粉機器人系統，授粉機器人能自動化辨識與計算花朵座標並執行授粉作業，同時研究中彙整出授粉機器人所需的技術與系統架構，至此確立了四軸飛行器應用於花朵授粉自動化的可行性。

關鍵字

授粉機器人；四軸飛行器；室內定位；自動控制

並列摘要

In recent years, the issue of Colony Collapse Disorder has received much attention; besides, pollinating birds and mammals also face the crisis of extinction in the same time. There are plenty of studies trying hard to clarify the above problem, and researchers also begin to look for alternative methods of pollination. Since pollinators will be prevented from entering greenhouses, artificial pollination or purchase of honey bees would be necessary to conduct the pollination work. While artificial pollination requires high labor force and purchase of honey bees has some limitations; there are very few studies regarding flying robots for pollination, especially those need physical contacts such as bionic pollination of insect and artificial pollination. Therefore, for the development of automation of pollination implemented on flying robots and exploring its possibility, this study developed an automated robotic system for pollination of flowers using a quadcopter equipped with video module and lightweight pollinating device. The robotic system can be divided into four subsystems, including pollination robotic system, image positioning system, decision control system and flower identification system. In the process of pollination, we first used depth camera to locate the flying robot, and guided the robot to target positions by using automatic control theory. Flowers were then identified by video module on the robot and it would be guided above the flower. After that, the position of the flower would then be determined by calculating the regression equation between the area of the flower and the height of the robot. Finally, the pollination duty could be accomplished by using pollinating device to touch the flower. In the positioning of the flying robot, the RMSE (Root Mean Square Error) values of X, Y and Z-axes were 4.96, 2.97 and 4.29 cm; and the RMSE value of absolute distance was 7.27 cm. The results of positioning indicated that the precision of control could reach an acceptable level. In terms of operation time, it required 24.71 seconds to finish the pollination work from the start of navigation of the flying robot to the contact of the flower. In summary, this study has successfully developed an automatic robotic system for pollination which is able to identify the position of flower and execute the pollination work. In the same time, the required technologies and the architecture of the system are integrated in this study to confirm the possibility of the automation of pollination of flower implemented on flying robots.